Classifier training methods, usage methods, devices, equipment and storage media

By calculating the uncertainty of multimedia samples and training a classifier, the problem of poor generalization performance caused by the differences in training sample subgroups is solved, and higher classifier generalization ability and test set accuracy are achieved.

CN115272797BActive Publication Date: 2026-06-30TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
TENCENT TECHNOLOGY (SHENZHEN) CO LTD
Filing Date
2022-07-29
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing technologies fail to effectively balance the differences in training sample subgroups provided by different medical centers when training classifiers, resulting in poor generalization performance of the classifiers when there are large differences in the distribution of subgroups in the test set.

Method used

By calculating the uncertainty of multimedia samples and training a classifier based on multiple multimedia samples, their labels, and uncertainties, the concept of sample subgroups in the training set is removed, and the performance impact of each sample on the classifier is recalculated.

Benefits of technology

This improved the classifier's generalization ability, making it perform more consistently and accurately on test sets across different medical centers, and reduced its dependence on sample subgroup distribution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115272797B_ABST
    Figure CN115272797B_ABST
Patent Text Reader

Abstract

This application discloses a training method, usage method, apparatus, device, and storage medium for a classifier, belonging to the field of artificial intelligence. The method includes: acquiring multiple multimedia samples and their labels from a training set; iteratively inputting the multiple multimedia samples into the classifier to obtain multiple prediction results; for a first multimedia sample among the multiple multimedia samples, generating a first uncertainty based on the multiple prediction results obtained from the multiple iterations of the first multimedia sample and its label, wherein the first uncertainty measures the ease with which the classifier accurately predicts the first multimedia sample; and training the classifier based on the multiple multimedia samples, their labels, and the uncertainties. This method improves the generalization ability of the trained classifier.
Need to check novelty before this filing date? Find Prior Art